dsh-super-advisor

DSH SuperAdvisor tackles a specific problem: you want your main model to handle routine work with a lightweight, cost-effective release, yet lack a safe, controllable channel to consult a stronger model when things get difficult. The plugin standardizes that channel through the ask_advisor tool call. The main model submits a question, goal, constraints, attempted actions, and up to eight evidence items; after approval, only the selected text is sent to the advisor model, and the returned advice is verified and acted upon by the main model rather than executed automatically.

Core capabilities center on controllability and auditability. Approval modes cover manual confirmation, independent AI review (an extra model call), main-model tagging (no extra call), and full auto-approval. Every mode is bound to an immutable snapshot, so any configuration change immediately invalidates pending grants. Each request invokes the DSH stream interface exactly once, with no automatic retries or provider failover. A send marker is persisted before the call starts, so re-entering the same task either reuses the stored result or refuses to send again.

This plugin suits developers running local models who want on-demand access to a stronger cloud model, particularly the Flash-plus-Pro cost-reduction setup. If you need fine-grained control over which text may leave your machine, a complete approval audit trail, and a plugin that respects the existing file and command permission system of DSH without bypassing it, SuperAdvisor is a focused and restrained choice.